David E. Culler University of California, Berkeley August 2, 2011
What can we put on the table? 2 8/2/2011CPS-PI-11
Sustainability? “Sustainable development should meet the needs of the present without compromising the ability of future generations to meet their own needs” Our Common Future, World Commission on Environment and Development, United Nations, /2/2011CPS-PI-11
Quantifying it - California Law AB 32 Reduce GHG emissions to 1990 levels by 2020 Governor’s executive order S-3-05 (2005) 80% reduction below 1990 levels by 2050 Renewable Portfolio Standard 33% renewables by 2020, 20% biopower procurement 480 => 80 mmT CO2e in 40 years Population expected to grow from 37 => 55 million Economic growth 4 8/2/2011CPS-PI-11
5 8/2/2011CPS-PI-11
What it amounts to … 6 8/2/2011CPS-PI-11
Can we do this? 7 8/2/2011CPS-PI-11
But 8 8/2/2011CPS-PI-11
My Take … We can’t build our way to sustainability The discussion is focused on the Physical New Windows, Materials, Motors, Biofuels, … Need to make the best of what we have, or will have, and how we use it… This takes observation, intelligence and control Demonstrable CPS technology on the table Efficiency & Supply-Following Loads 10 years to innovate, 30 years to scale 9 8/2/2011CPS-PI-11
What’s on the Table? – buildings 10 8/2/2011CPS-PI-11 Deployed Demonstrated Development Research Concept
4 Part GHG Reduction Plan Efficiency 11 8/2/2011CPS-PI-11
4 Part GHG Reduction Plan Efficiency Electrify 12 8/2/2011CPS-PI-11
4 Part GHG Reduction Plan Efficiency Electrify Decarbonize the electricity Decarbonize the fuel 13 8/2/2011CPS-PI-11
4 Part GHG Reduction Plan Efficiency Electrify Decarbonize the electricity Decarbonize the fuel 14 8/2/2011CPS-PI-11
All required for even 60% reduction but still fall short of 80% 8/2/ CPS-PI-11
Efficiency 16 8/2/2011CPS-PI-11
Electrification 17 8/2/2011CPS-PI-11
Low Carbon Electricity Options 18 8/2/2011CPS-PI-11
Build Rate 19 8/2/2011CPS-PI-11
The Renewables are there 20 8/2/2011CPS-PI-11 43% agriculture, 3.4% urbanized
The Problem: Supply-Demand Match 21 Baseline + Dispatchable Tiers Oblivious Loads Transmission Generation Demand Distribution 8/2/2011CPS-PI-11
Load-following Supply (?) 8/2/2011CPS-PI Source: EEX spot prices. Growing proportion of renewables leads to higher price volatility. October 2008 to March 2010: >90 hours with negative prices; highest price reached: +€500/MWh, lowest -€500/MWh Daily minimum price (indicated in red when negative) Daily maximum price €/MWh
… and in CA 8/2/2011CPS-PI-11 23
Zero Emissions Load Balancing (ZELB) Just the emissions from the natural gas to firm the 33% renewables exceeds GHG target Even with 50% with natural gas & 50% with some yet-to-exist storage 24 8/2/2011CPS-PI-11
We Understand Supply 25 8/2/2011CPS-PI-11
Not demand 26 8/2/2011CPS-PI-11
Limits to Renewable Penetration Variability, Intermittency of Supply Visibility into Availability of Supply Ability of Loads to Adapt Algorithms and Techniques for Reactive Load Adaptation Capability of the Infrastructure to maintain the match 8/2/2011CPS-PI-11 27
CPS and the 4 Part GHG Reduction Plan Efficiency Electrify Decarbonize the electricity Decarbonize the fuel 28 Monitoring, Analysis, Modeling, Waste Elimination, Power Proportionality, Optimal Control Intelligence, Communication, adaptation in Everything ZELB. Supply-Following Loads, Energy SLA, Cooperative Grid 8/2/2011CPS-PI-11
CPS Technology … National-scale Physical Information Service Data collection, streaming, archiving, querying Modeling, Analysis, Control Representation, metadata, life-cycle Use others’, contribute yours 29 8/2/2011CPS-PI-11
Some starts 30 8/2/2011CPS-PI-11
sMAP: Uniform Access to Diverse Physical Information 8/2/2011CPS-PI Electrical Weather Geographical Water Environmental Structural Actuator Occupancy sMAP Modeling Visualization Continuous Commissioning Control Personal Feedback Debugging Storage Location Authentication Actuation Applications Physical Information REST API HTTP/TCP … JSON Objects
Internet Cell phone sMAP Gateway sMAP Modbus RS-485 sMAP sMAP Gateway EBHTTP / IPv6 / 6LowPAN Wireless Mesh Network sMAP Edge Router Temperature/PAR/TSR Vibration / Humidity AC plug meter Light switch Dent circuit meter Proxy Server EBHTTP Translation California ISO sMAP Gateway sMAP Resources Applications Google PowerMeter Weather sMAP Every Building Database sMAP ecosystem 8/2/2011CPS-PI-11 32
Cory CEC B2G Testbed 8/2/2011CPS-PI Whole Bldg MCL equip MCL infra MCL vac servers DOP HVAC Central vent office HVAC inst Lab 199 HVAC Plug loads Lighting Parking Lot
DOE MELS => Appliance Energy 8/2/2011CPS-PI LBNL Bldg of 1200 loads
DOE/UCB/Siemens Auto Demand Response 8/2/ Sutardja Dai Hall CPS-PI-11
BACNet => sMAP Sensors, Actuators Controllers Modem Siemens P2 over RS-485 Apogee Server Remote Login BACnet/IP Gateway sMAP Internet Sutardja Dai Hall 8/2/ CPS-PI-11
Data from 1 Modern Building 1358 control settings Set points, Relays (lights, pumps, etc), Schedules 2291 meters/sensors Power (building, floor, lights, chiller, pumps, etc) Current, voltage, apparent, real, reactive, peak Temp (rooms, chilled water, hot water) Air volume Alarms, Errors 2165 control outputs Dampers, valves, min/max flow, fan speed, PID parameters 72 other 4+ million Commercial, 110+ million Residential 8/2/ CPS-PI-11
Current sMAP demography NameSensor TypePhysical LayerSense PointsChannels Cory Hall SubmeteringDent Three-PhaseModbus + Ethernet Cory Hall MeteringION and pQube MetersHTTP/Ethernet3150 Cory Lab TemperatureTelosB Ethernet48 Cory Lab MachinesACme Ethernet816 Cory Chilled WaterHeatXModbus + Ethernet111 Cory WeatherVaisala WXT520SDI-12 + Ethernet111 Soda Hall Sun BlackboxFan speed; environmentalHttp/Ethernet1084 Soda Lab MachinesACme Ethernet4080 Soda Lab PanelVeris E30Modbus + Ethernet142 Soda SCADA DataBarrington controlsRS-485/various“1”1670 LBNL Building 90Acme Ethernet Campus Power DataObvius Aquisuite; variousXML/HTTP/Ethernet4100 Citris SDH BACnetSiemens ApogeeBACnet/IP + RS-485“1”600+ Brower BuildinbgJohnsons ControlBACnet“1”1500 8/2/ CPS-PI-11
insert resample aggregate query streaming pipeline Page CacheLock Manager Key-Value Store Storage Alloc. Time-series Interface Bucketing RPC Compression readingdb SQL Storage mapper MySQL A Stream Engine Dent circuit meterDent circuit meter sMAP ~300k pts/sec 8/2/ CPS-PI-11
Scaling Out in the Cloud 8/2/2011CPS-PI Stream Management System Data Processing Data Source Data Sink Real Time Monitoring HBaseHive HDFS Map-Reduce Hadoop Interactive Workloads Analytics Workloads Columnar Data Storage Query Translator SQL MR
15,000 pts and growing… 8/2/2011CPS-PI-11 41
CPS Technology National-Scale Physical Info Service Software foundations, platforms and solutions for Energy Efficiency and Agility 42 8/2/2011CPS-PI-11
Stages of Energy Effectiveness Waste Not Do Nothing Well !!! Power Proportionality Peak Performance : Power => Safety Optimize Partial Load - from nothing to peakl Sculpting Identify the energy slack and utilize it Negotiated Grid / Load / Human Interaction Plan, Forecast, Negotiate, Manage 8/2/2011CPS-PI-11 43
8/2/ Lighting Servers PDUs, CRACs Our Buildings HVAC IT and Plug Load CPS-PI-11
8/2/ Power- Proportional Buildings ? 950 KW 1150 KW Min = 82% of Max Cory Hall: Office + Semiconductor + IT CPS-PI-11
8/2/ Power- Proportional Buildings ? 1.45 MW 2.02 MW Min = 72% of Max Stanley Hall: Office + BioScience - 13 NMRs CPS-PI-11
8/2/ Power- Proportional Buildings ? 62 KW 202 KW Min = 31% of Max LeConte Hall: Office CPS-PI-11
Re-Flash the HVAC … 48 Whole Bldg inst Lab 199 HVAC 8/2/2011CPS-PI-11 LoCal + ActionWebs
Learning-based Model Predictive Control Mathematical model from Newton’s law of cooling Model identified using semi-parametric regression dT/dt = -k r T - k c u(t) + k w w(t)) + q(t) change in temperature over time time constant of room AC cooling weather heating from occupants and equipment Temperature: Experimental (blue) Simulated (red) Heating from occupants and equipment 8/2/ (Aswani, Master, Taneja, Culler, Tomlin, Proc. of IEEE, 2011) CPS-PI-11
Living Lab… LBMPC saved 30% energy when hotter outside 70% energy when cooler outside Standard MPC was inconsistent Two-position control was inefficient 50 Method Measured Energy Estimated Energy Tracking Error Temperature Variation Average External Load Two-Position Control Experiment LBMPC23.6kWh0.75°C0.13°C11.0°C Linear MPC30.5kWh0.62°C0.30°C11.0°C Two-Position32.6kWh35.1kWh0.61°C0.20°C11.0 °C LBMPC Experiment LBMPC11.8kWh13.3kWh0.86°C0.17°C7.2°C Linear MPC8.6kWh0.93°C0.21°C7.2°C Two-Position34.5kWh0.55°C0.19°C7.2°C 8/2/2011CPS-PI-11
LBMPC to Minimize Energy… Estimates heating load using model Best control that considers estimated occupancy (Aswani, Master, Taneja, Culler, Tomlin, 2011, submitted) 51 Cost AC Constraints Temperature Constraints Temperature Dynamics 8/2/2011CPS-PI-11
… or optimize to follow the supply 52 Generation Transmission Distribution Load VPG 8/2/2011CPS-PI-11
at Building-Scale, and beyond 53 Temperature Blue – Measured; Red - Simulated Heating from occupants, equipment, etc. 8/2/2011CPS-PI-11
Making Sense out of Sensors 54 °°°°°° °°°°°° °°°°°° 8/2/2011CPS-PI-11
Supply Air Fan VFD 55 8/2/2011CPS-PI-11
What’s really going on? Bring comfortable air from outside Cool it down till its really cold Push it out everywhere thru VAVs that are at minimum opening Reheat it to set point So the empty rooms will be comfortable This is going on everywhere! And we supply perfect, precious energy to do it! 56 8/2/2011CPS-PI-11
And IT is no better … Core i7 Atom 333 Westmere 4/12/11 57 CPS 2011
Supply-Following Computational Loads 8/2/2011CPS-PI IPS Requests Power Background Processing (shiftable) Controllable Storage QoS (fidelity & latency) Availability Forecasts
Non-dispatchable, variable supply Power proportional, grid-aware loads NREL Western Wind and Solar Integration Study Dataset Pacheco wind farm Scientific computing cluster Energy-Availability Driven Scheduling 8/2/2011CPS-PI-11 59
Scheduling & Energy Storage 8/2/ CPS-PI-11
CPS Technology National Physical Info Service Software foundations for Energy Efficiency and Agility Infrastructure for Energy Innovation 61 8/2/2011CPS-PI-11
Personalized Automated Lighting Control Three controllable ballasts per fixture ~5 zones per floor BACnet sMAP MySQL Python Django Python Control Process HTTP Gateway 8/2/ CPS-PI-11
Real Energy Savings 8/2/2011CPS-PI-11 63
From the agony of load-following … 64 8/2/2011CPS-PI-11
to Co-operative Energy Mgmt in a Cyber-Physical Grid 65 Load IPS Source IPS energy subnet Intelligent Power Switch Monitor, Model, Mitigate Deep instrumentation Power proportional Design Energy-Agile Control Shifting, Scheduling, Adaptation Forecasting Tracking Market Availability Pricing Planning 8/2/2011CPS-PI-11
Thanks 8/2/2011CPS-PI-11 66